How to use "model_parameter_search.create" with a custom neural network?

User 5238 | 6/3/2016, 11:33:55 PM

I am using graphlab.deeplearning.create to create an initial neural network, and I want to perform a model parameter search using model_parameter_search.create, but when I specify the initial network in the params dictionary like this:

params = {"target": "FINAL_STATE", "network" : self.init_net, "features" : self.features, "max_iterations" : 8, "learning_rate" : 0.01 , "validation_set" : None} job = graphlab.model_parameter_search.create(folds, graphlab.neuralnet_classifier.create, params) I got: Unrecognized search value for parameter network If you could point me to the right to do this search?

Comments

User 1190 | 6/7/2016, 4:51:12 AM

Can you try put the "self.init_net" into a list, e.g.:

params = {"target": "FINAL_STATE", "network" : [self.init_net], "features" : self.features, "max_iterations" : 8, "learning_rate" : 0.01 , "validation_set" : None} job = graphlab.model_parameter_search.create(folds, graphlab.neuralnet_classifier.create, params)